In this paper, we validate the simple water balance model incorporated in the Carnegie Ames Stanford Approach (CASA) model with hourly soil moisture observations at a semi-arid site in the Ordos plateau of Inner Mongolia. The CASA model is widely used for estimating large-scale net primary productivity (NPP) based on satellite data. The results indicated that the soil moisture submodel of the CASA model did not simulate absolute values and seasonal changes observed at the Ordos site in Inner Mongolia. In order to realistically estimate soil moisture, we considered the following two points: (1) the time interval of the model is set to be daily instead of monthly and (2) a new scheme of evaporation efficiency for daily calculation that produces a wide range of evaporation variation. The soil-moisture model proposed by the present study will be useful for estimating precise NPP for semi-arid land, based on process models.